• Motivation (~ 1 minute)
• Algorithm in brief (~ 5 minutes)
• Results (~ 5 minutes)
• Discussion (~ 5 minutes)

• This is a preprint
• I am not a historian / anthropologist / archaeologist
• Slides are available at https://github.com/Chris1221/eghop_270720

• Sequential Monte Carlo for the Sequentially Markovian Coalescent
• Infer demographic parameters (effective population size, directional migration) over a number of piecewise constant epochs
  • Similar to MSMC(2) in this regard, but we control the epochs.
• Python package
  • Install with conda
  • https://github.com/luntergroup/smcsmc

• Simulate ancestral recombination graphs sequentially across the genome with parameters \(\theta\)
  • Sequential coalescent with recombination model (scrm))
• Using information about variation, asign a weight \(w_i\) to each of the particles (histories).
• As you move along the genome, resample if the collection of particles becomes too similar (effective sample size gets too low).
• Use variational Bayes to update the parameter estimtes \(\theta\) until convergence

ESS = Effective sample size (of the particle population)

• Pairs of individual WGS samples from Simons Genome Diversity Panel (SGDP) and Human Genome Diversity Project (HGDP)
  • HGDP samples experimentally phased, SGDP statistically phased.
• Simulations done in SCRM with known demographic parameters
• \(D\) statistics (a la Patterson 2012) in admixr (highly recommend)
  • Ancient samples include Vindija and Altai Neanderthals
  • Many more from the Reich Human Origins dataset
• All experiments were replicated three times.

IMF = Integrated migration fraction, or the total proportion replaced over a given time period.

In order, simulated Eurasia to Africa migration, bidirectional, and Africa to Eurasia migration. Many more scenarios in the supplemental material.

MSMC2 run with default parameters.